diff --git a/litellm/llms/databricks/chat/transformation.py b/litellm/llms/databricks/chat/transformation.py
index 538904b34e6..063462287f8 100644
--- a/litellm/llms/databricks/chat/transformation.py
+++ b/litellm/llms/databricks/chat/transformation.py
@@ -387,7 +387,15 @@ class DatabricksConfig(DatabricksBase, OpenAILikeChatConfig, AnthropicConfig):
             )  # unsupported for claude models - if json_schema -> convert to tool call
 
         if "reasoning_effort" in non_default_params and self._uses_anthropic_thinking_param(model):
-            reasoning_effort_value: Final = non_default_params.get("reasoning_effort")
+            reasoning_effort_value = non_default_params.get("reasoning_effort")
+            # Accept both the string form ("low") and the dict form
+            # ({"effort": "low", "summary": "concise"}). The Responses->Chat
+            # parser keeps the full dict whenever `summary` is set (#28196),
+            # so a dict is the standard shape Responses-bridge callers send.
+            # Same coercion the direct Anthropic and Bedrock Converse
+            # adapters already do.
+            if isinstance(reasoning_effort_value, dict):
+                reasoning_effort_value = reasoning_effort_value.get("effort")
             mapped_thinking: Final = AnthropicConfig._map_reasoning_effort(
                 reasoning_effort=reasoning_effort_value,
                 model=model,
diff --git a/tests/unit/llms/databricks/chat/test_databricks_chat_transformation.py b/tests/unit/llms/databricks/chat/test_databricks_chat_transformation.py
index 52bb89fed5a..888c03a46b4 100644
--- a/tests/unit/llms/databricks/chat/test_databricks_chat_transformation.py
+++ b/tests/unit/llms/databricks/chat/test_databricks_chat_transformation.py
@@ -544,7 +544,7 @@ def test_map_openai_params_upgrades_legacy_thinking_on_adaptive_only_claude(
     assert mapped.get("output_config") == expected_output_config
 
 
-def _map_reasoning_effort(model: str, reasoning_effort: str):
+def _map_reasoning_effort(model: str, reasoning_effort: str | dict):
     return DatabricksConfig().map_openai_params(
         non_default_params={"reasoning_effort": reasoning_effort},
         optional_params={},
@@ -578,6 +578,22 @@ def test_unmapped_claude_endpoint_still_translates(_use_local_model_cost_map):
     assert "reasoning_effort" not in params
 
 
+@pytest.mark.parametrize(
+    "model",
+    ["databricks-claude-3-7-sonnet", "my-claude-serving-endpoint"],
+)
+def test_reasoning_effort_dict_is_coerced(_use_local_model_cost_map, model):
+    """The Responses->Chat parser keeps `reasoning_effort` as a dict whenever
+    `summary` is set (#28196). Without coercion the dict reaches
+    `_map_reasoning_effort` and no thinking payload is produced."""
+    params = _map_reasoning_effort(model, {"effort": "low", "summary": "concise"})
+    assert params.get("thinking") == {
+        "type": "enabled",
+        "budget_tokens": DEFAULT_REASONING_EFFORT_LOW_THINKING_BUDGET,
+    }
+    assert "reasoning_effort" not in params
+
+
 def test_gemini_2_5_low_translates_to_thinking_budget(_use_local_model_cost_map):
     params = _map_reasoning_effort("databricks-gemini-2-5-flash", "low")
     assert params.get("thinking") == {